DOI: 10.1097/CCM.0000000000000078

,  

,  

PMID: 24346549

Issn Print: 0090-3493

Publication Date: 2014/01/01


Print

The authors reply

Peter Pickkers; Niels Peek; Nicolet de Keizer

    loading  Checking for direct PDF access through Ovid

Excerpt

We thank Xue et al (1), Lin (2), and Claessen et al (3) for their interest in our study (4). Xue et al (1) highlight the possibility of uncontrolled confounding in an observational study such as ours, an important issue that we also mentioned in the discussion of our article. We like to emphasize that in the previous meta-analyses on this subject (see [4]), results were not corrected for any other prognostic covariate, whereas in our database, it was possible to do so for most known risk factors, but obviously not all. In our view, registration of over a 100 patient characteristics and with over 90% of all Dutch ICUs participating, the Netherlands Intensive Care Evaluation (NICE) database allows us to explore associations between patient characteristics and outcome in a sufficiently sophisticated manner.
To specifically address the issues raised by Xue et al (1): the Simplified Acute Physiology Score II includes 17 variables: 12 physiology variables, age, type of admission (scheduled surgical, unscheduled surgical, or medical), and three underlying disease variables (AIDS, metastatic cancer, and hematologic malignancy) (5). Its performance in the Dutch population is comparable to Acute Physiology and Chronic Health Evaluation II and IV, but data are available for a longer time period (6). We did not correct for chronic diagnoses, such as heart failure, as it might be the consequence, not the cause, of obesity. This is different from chronic diagnosis such as AIDS or cancer, where weight loss is clearly related to the chronic diagnosis and correction is justified. Second, transfusion data are, unfortunately, not collected in the NICE database, but in our view, it appears unlikely that physicians would use different transfusion triggers based on the body mass index (BMI) of the patient. Also ethnicity is not collected, but with 3–4% blacks in The Netherlands (7), it unlikely represents a potential bias of relevance. Finally, all residents in the Netherlands are obliged by law to purchase health insurance coverage, resulting in only 0.8% of people without insurance (8), also suggesting that the relevance of possible bias induced by uninsured patients is limited. The authors are correct that we do not have details available about the therapies the patients receive during their stay in the ICU.
Lin (2) highlights the possibility that the obesity paradox might not be present in specific subgroups of critically ill patients. We agree that it appears to be plausible that the presence of the obesity paradox is related to the diagnosis. In our article, we wished to demonstrate the absence/presence of the obesity paradox in ICU patients using our database of over 150,000 patients to have sufficient statistical power. The study by Lin (2) refers to apply dichotomous analyses (at a BMI of 30 kg/m2), and in view of the inverse J-shape outcome, this may dilute putative beneficial associations between BMI and outcome. For example, in the H1N1 study cited (3,059 patients), an increased risk of dying was found in patients with a BMI more than 40, whereas for patients with a BMI more than 30, this did not reach statistical significance. In the study in 43 patients with tuberculous meningitis, BMI is not mentioned as a covariate, and the study would be underpowered to detect a possible effect of BMI on outcome.
    loading  Loading Related Articles